*: add MVCC version ratio to slow log metrics (#44897) (#47645)

close #44896

Author: ti-chi-bot

executor/adapter.go

@@ -1606,6 +1606,9 @@ func (a *ExecStmt) LogSlowQuery(txnTS uint64, succ bool, hasMoreResults bool) {
 			executor_metrics.TotalQueryProcHistogramGeneral.Observe(costTime.Seconds())
 			executor_metrics.TotalCopProcHistogramGeneral.Observe(execDetail.TimeDetail.ProcessTime.Seconds())
 			executor_metrics.TotalCopWaitHistogramGeneral.Observe(execDetail.TimeDetail.WaitTime.Seconds())
+			if execDetail.ScanDetail != nil && execDetail.ScanDetail.ProcessedKeys != 0 {
+				executor_metrics.CopMVCCRatioHistogramGeneral.Observe(float64(execDetail.ScanDetail.TotalKeys) / float64(execDetail.ScanDetail.ProcessedKeys))
+			}
 		}
 		var userString string
 		if sessVars.User != nil {

executor/metrics/metrics.go

@@ -43,6 +43,7 @@ var (
 	TotalQueryProcHistogramGeneral  prometheus.Observer
 	TotalCopProcHistogramGeneral    prometheus.Observer
 	TotalCopWaitHistogramGeneral    prometheus.Observer
+	CopMVCCRatioHistogramGeneral    prometheus.Observer
 	TotalQueryProcHistogramInternal prometheus.Observer
 	TotalCopProcHistogramInternal   prometheus.Observer
 	TotalCopWaitHistogramInternal   prometheus.Observer
@@ -137,6 +138,7 @@ func InitMetricsVars() {
 	TotalQueryProcHistogramGeneral = metrics.TotalQueryProcHistogram.WithLabelValues(metrics.LblGeneral)
 	TotalCopProcHistogramGeneral = metrics.TotalCopProcHistogram.WithLabelValues(metrics.LblGeneral)
 	TotalCopWaitHistogramGeneral = metrics.TotalCopWaitHistogram.WithLabelValues(metrics.LblGeneral)
+	CopMVCCRatioHistogramGeneral = metrics.CopMVCCRatioHistogram.WithLabelValues(metrics.LblGeneral)
 	TotalQueryProcHistogramInternal = metrics.TotalQueryProcHistogram.WithLabelValues(metrics.LblInternal)
 	TotalCopProcHistogramInternal = metrics.TotalCopProcHistogram.WithLabelValues(metrics.LblInternal)
 	TotalCopWaitHistogramInternal = metrics.TotalCopWaitHistogram.WithLabelValues(metrics.LblInternal)

metrics/grafana/tidb.json

@@ -814,31 +814,44 @@
           "pointradius": 5,
           "points": false,
           "renderer": "flot",
-          "seriesOverrides": [],
+          "seriesOverrides": [
+            {
+              "$$hashKey": "object:211",
+              "alias": "/mvcc_ratio/",
+              "yaxis": 2
+            }
+          ],
           "spaceLength": 10,
           "stack": false,
           "steppedLine": false,
           "targets": [
             {
-              "expr": "histogram_quantile(0.90, sum(rate(tidb_server_slow_query_process_duration_seconds_bucket{k8s_cluster=\"$k8s_cluster\", tidb_cluster=\"$tidb_cluster\", instance=~\"$instance\"}[1m])) by (le,sql_type))",
+              "expr": "histogram_quantile(0.90, sum(rate(tidb_server_slow_query_process_duration_seconds_bucket{k8s_cluster=\"$k8s_cluster\", tidb_cluster=\"$tidb_cluster\", instance=~\"$instance\", sql_type=\"general\"}[1m])) by (le,sql_type))",
               "format": "time_series",
               "intervalFactor": 2,
-              "legendFormat": "all_proc_{{sql_type}}",
+              "legendFormat": "all_proc",
               "refId": "A"
             },
             {
-              "expr": "histogram_quantile(0.90, sum(rate(tidb_server_slow_query_cop_duration_seconds_bucket{k8s_cluster=\"$k8s_cluster\", tidb_cluster=\"$tidb_cluster\", instance=~\"$instance\"}[1m])) by (le,sql_type))",
+              "expr": "histogram_quantile(0.90, sum(rate(tidb_server_slow_query_cop_duration_seconds_bucket{k8s_cluster=\"$k8s_cluster\", tidb_cluster=\"$tidb_cluster\", instance=~\"$instance\", sql_type=\"general\"}[1m])) by (le,sql_type))",
               "format": "time_series",
               "intervalFactor": 2,
-              "legendFormat": "all_cop_proc_{{sql_type}}",
+              "legendFormat": "all_cop_proc",
               "refId": "B"
             },
             {
-              "expr": "histogram_quantile(0.90, sum(rate(tidb_server_slow_query_wait_duration_seconds_bucket{k8s_cluster=\"$k8s_cluster\", tidb_cluster=\"$tidb_cluster\", instance=~\"$instance\"}[1m])) by (le,sql_type))",
+              "expr": "histogram_quantile(0.90, sum(rate(tidb_server_slow_query_wait_duration_seconds_bucket{k8s_cluster=\"$k8s_cluster\", tidb_cluster=\"$tidb_cluster\", instance=~\"$instance\", sql_type=\"general\"}[1m])) by (le,sql_type))",
               "format": "time_series",
               "intervalFactor": 2,
-              "legendFormat": "all_cop_wait_{{sql_type}}",
+              "legendFormat": "all_cop_wait",
               "refId": "C"
+            },
+            {
+              "expr": "histogram_quantile(0.90, sum(rate(tidb_server_slow_query_cop_mvcc_ratio_bucket{k8s_cluster=\"$k8s_cluster\", tidb_cluster=\"$tidb_cluster\", instance=~\"$instance\", sql_type=\"general\"}[1m])) by (le,sql_type))",
+              "format": "time_series",
+              "intervalFactor": 2,
+              "legendFormat": "mvcc_ratio",
+              "refId": "D"
             }
           ],
           "thresholds": [],
@@ -861,6 +874,7 @@
           },
           "yaxes": [
             {
+              "$$hashKey": "object:119",
               "format": "s",
               "label": null,
               "logBase": 2,
@@ -869,6 +883,7 @@
               "show": true
             },
             {
+              "$$hashKey": "object:120",
               "format": "short",
               "label": null,
               "logBase": 1,

metrics/metrics.go

@@ -204,6 +204,7 @@ func RegisterMetrics() {
 	prometheus.MustRegister(TotalQueryProcHistogram)
 	prometheus.MustRegister(TotalCopProcHistogram)
 	prometheus.MustRegister(TotalCopWaitHistogram)
+	prometheus.MustRegister(CopMVCCRatioHistogram)
 	prometheus.MustRegister(HandleSchemaValidate)
 	prometheus.MustRegister(MaxProcs)
 	prometheus.MustRegister(GOGC)

metrics/server.go

@@ -56,6 +56,7 @@ var (
 	TotalQueryProcHistogram         *prometheus.HistogramVec
 	TotalCopProcHistogram           *prometheus.HistogramVec
 	TotalCopWaitHistogram           *prometheus.HistogramVec
+	CopMVCCRatioHistogram           *prometheus.HistogramVec
 	MaxProcs                        prometheus.Gauge
 	GOGC                            prometheus.Gauge
 	ConnIdleDurationHistogram       *prometheus.HistogramVec
@@ -256,6 +257,15 @@ func InitServerMetrics() {
 			Buckets:   prometheus.ExponentialBuckets(0.001, 2, 28), // 1ms ~ 1.5days
 		}, []string{LblSQLType})
 
+	CopMVCCRatioHistogram = NewHistogramVec(
+		prometheus.HistogramOpts{
+			Namespace: "tidb",
+			Subsystem: "server",
+			Name:      "slow_query_cop_mvcc_ratio",
+			Help:      "Bucketed histogram of all cop total keys / processed keys in slow queries.",
+			Buckets:   prometheus.ExponentialBuckets(0.5, 2, 21), // 0.5 ~ 262144
+		}, []string{LblSQLType})
+
 	MaxProcs = NewGauge(
 		prometheus.GaugeOpts{
 			Namespace: "tidb",